Instantaneous Variable Bitrate Visualization between MP3, FLAC, AAC, and Opus Audio Codecs

[@abunk8691]

Its interesting that much FLAC scales its bitrate with the volume of the audio in a VU meter way that shows dynamics of the audio which AAC LC also does to a much lesser extent. Though Opus by comparison doesn't drop the bitrate much on lower volume sections like at the very beginning and end of the song. They all got peaky with the bitrates with the higher frequency sounds like drum hi-hats and some piano parts, but not as much with the brass like instruments (not sure if its trumpets) like at 0:28. Opus spiking more with the drums (like another commenter mentioned) is interesting though, but does kinda make sense since Opus is a codec built for encoding speech and not so much for music unlike FLAC. The framerate of the bars initially bothered me until I remembered how much of a pain it was to smoothen visualizers of audio levels when I was playing with Rainmeter years ago. Considering there's 4 stream of audio whose variable bitrate is being monitored and represented as bars in real time I would assume this was quite CPU intensive to do.

[@KYLXBN]

Awesome analysis! Opus is indeed built for encoding speech (if we can say that) but it's not "specialized" for speech. It just prioritizes speech as much as music. It's actually two codecs in one--SILK which specializes in voice, and CELT for general music. It's impressive--on some bitrates, both can be active at the same time, and Opus can also switch between them freely when it detects that a part contains only voice or when a podcast suddenly switches to music. But for higher bitrates such as this one, only CELT is active. The framerate is actually intentionally that way! That's how fast each codec changes bitrates. I could smooth it out--but then, it won't represent the reality of how fast the bitrate actually changes ^_^ I'm glad you found it as interesting as I did!

[@phazonclash]

Opus is a phenomenal lossy format.

[@KYLXBN]

Indeed it is! ^_^

[@myoukochou]

@@KYLXBN It really is incredible... hmm, I see we still could do a bit better in the encoder for the quiet bit at the start! :)

[@KYLXBN]

@@myoukochou I was wondering about that too! When other codecs just gave up encoding silence, but Opus just goes and "well, time to encode -60 dB 1-bit noise"

[@MarkNeri]

Was obsessed with the original version of this song especially because they managed to cram it into a Super Famicom cart. But I prefer this version on PS1. Goosebumps the first time I heard it...and perhaps every time

[@Kasatome]

Really cool to see how they differ on a graph like this. I only started using FLAC once I didn't have to worry about space. I was always the outcast when choosing an audio format, which was always WMA.

[@KYLXBN]

WMA is indeed outcast! As a Linux user, I used to store my collection in FLAC, and the portable copy would be in Vorbis. Then, around a couple years ago, I switched to WavPack and Opus :)

[@ToufouMaster]

isn't the green and red inverted?

[@KYLXBN]

Oh, shoot, you're right!

[@ToufouMaster]

@@KYLXBN we still understand but that's just quite confusing. great video as always

[@KYLXBN]

@ToufouMaster Thank you! I hope it was informative.

[@Littlepetfrog]

I loved it! ​@@KYLXBN

[@thafff]

Just to preshoot the question: Yukari Yoshida - Yume wa Owaranai ～Kobore Ochiru Toki no Shizuku～ Nice song, and no thanks to OP for reminding me its existence, because I'm compelled to acquiring the CD :3c

[@ColonDee.]

Tales of Phantasia ❤

[@hexecuted]

Would be interesting to see waveform views for each type as it plays, but only showing the difference/xor between the original and encoded views to highlight the difference in audio reproduction.

[@KYLXBN]

Well, we could do that, but to be honest, that's not the right way to compare lossy codecs. It's a very long story, but the difference from the original waveform *cannot* tell you the quality of the sound. You don't judge the audio quality via its waveform because you don't listen to music via its waveform.

[@phosphor3617]

Kyle X-bone with another banger

[@Anton_Platonov]

Hm-m, the FLAC bar looks like a VU meter to me. Is it giving away how it works, just basically RLE for the unused bits of the dynamic range?

[@KYLXBN]

The louder the volume (and the noisier the data), the more bits FLAC needs to store the data, so yeah, I guess that's basically it :) Edit: to make it clear though, FLAC is not "just RLE", it's more complex than that.

[@myoukochou]

General Linear Predictive Coding with the residual coded by a Rice-Golomb code; the residual will require more bits for noisy bits, like kicks and snares, so that's why you're seeing that.

[@multicoloredwiz]

very excellent video!

[@KYLXBN]

Thank you, I hope it was informative.

[@MikeDawson1]

do you have the green and red colors backwards?

[@KYLXBN]

Yes, I did have it reversed by mistake :)

[@gungun974]

Tanks you for me discovered a new song !

[@AcrosArchive]

This is so cool!

[@rapsod1911]

And size of file comparison is... where? Is there anyway to compare outputs to original and see which codec is less "noisy"?

[@KYLXBN]

Unfortunately, the size is irrelevant to this experiment since the focus is bitrate. Still, I think you'd really rather compare bitrates than file sizes, for reasons I won't go too deep in. At 190kbps too, no codec would sound "noisy" since it's already at the point of "transparency"--when we can't tell between the original and the compressed version. For a codec to sound "noisy", we'd probably compare at around 32 to 64 kbps. There, we can definitely hear the differences :) And if you ask me about that, I'd say at 64 kbps, AAC wins by a slight margin over Opus, and that still depends on personal preference. Personally, I prefer Opus' raspy sound at 64 kbps compared to AAC's metalic sound at 64 kbps. Edit: What you're actually hearing in this video is the original FLAC (CD rip) audio (which KZbin will definitely convert to Opus so you're listening to Opus in the end)

[@KickCaesar]

@@KYLXBN Sometimes I like just compressing songs in different codecs at a low bitrate to listen what It'll sound like. It's kinda neat hearing the different artifacts of the codecs.

[@KYLXBN]

​@KickCaesar Me too, I used to do it a lot back when I was in high school, experimenting with various lossy codecs :)

[@aleksjenner677]

What's the song?

[@KYLXBN]

Tales of Phantasia (PSX) theme song, "Yume wa Owaranai" :)

[@CrushedAsian255]

What program did you use to make this?

[@KYLXBN]

I coded the program myself just for this video so it doesn't have a name.

[@CrushedAsian255]

@@KYLXBN what system did you use to get the individual frame sizes? Ffmpeg?

[@KYLXBN]

​@@CrushedAsian255 Yes :)

[@xcbrr50]

i know a guy who cut a whole documentary on ffmpeg

[@KYLXBN]

​@@xcbrr50That's crazy!

[@dottedboxguy]

interesting how opus seems to really dislike drums (super cool vid btw ! i always wanted to see such a comparison)

[@KYLXBN]

I believe most lossy codecs dislike instantaneous (snappy) sounds like maracas, and to a lesser extent, drums. It just messes up their MDCT compression so they have to allocate more bits during those instances to avoid degrading the sound quality. While not really shown in this demonstration, I think you might find it interesting that lossy codecs actually love noisy chaotic music (like metal) because it resembles noise and having the sound quality of noise degrade a bit wouldn't make it noticeable to us humans since degraded noise is still noise anyway ^_^

[@dottedboxguy]

@ that's really interesting yeah ! so much cool stuff in codecs (even tho i personally only use flac or wav as an audio enjoyer lol, when you have good audio equipment lossy compression gets quite noticeable, or at least lossless actually sounds better) edit : i also enjoy oscilloscope music which gets destroyed by any lossy codec even if really good

[@KYLXBN]

@@dottedboxguy Same here! My entire music collection is in WavPack (lossless) :) I also enjoy music and have the same complaints as you :)

[@LucasNascimentoDeAssisDias]

Could you show how Punch Out!! Training theme would sound in the enhanced NES Chip? I know it was a lot of years ago, but I would love to hear it. Great job!

[@KYLXBN]

Sure! I'll upload it as soon as I have free time :)

[@LucasNascimentoDeAssisDias]

@@KYLXBN Thank u! :D

[@martingerken7094]

FLAC has a much higher base rate, so the same relative change looks much bigger. Please scale graph!

[@KYLXBN]

It's all shown in linear scale :) This is not meant to antagonize FLAC or show it as inferior. But since FLAC does use more bits (due to its lossless nature), it's meant to be this way :) To make them look closer to each other, I'll have to remove FLAC from the graph. Maybe a follow-up video containing only lossless codecs is in order :)

[@krellion]

If you do that, I'd like to see Vorbis in FLAC's position. See how it compares to the newer Opus.